DETAILED ACTION
This action is written in response to the remarks and amendments dated 12/18/25. This action is made final. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
The Examiner acknowledges the terminal disclaimer filed 12/18/25, which was approved. Accordingly, the outstanding double patenting rejection is withdrawn.
The Applicants argue that the previous art of record does not anticipate or render obvious the claims as currently amended. The Examiner provides updated prior art rejections below necessitated by the current amendments.
Subject Matter Eligibility
In determining whether the claims are subject matter eligible, the examiner has considered and applied the 2019 USPTO Patent Eligibility Guidelines, as well as guidance in the MPEP chapter 2106. The examiner finds that each independent claim—when viewed as a whole—cannot be practically implemented as a mental process (eg due to performing vector operations within the embeddings layer of a neural network).
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103(a) which forms the basis for all obviousness rejections set forth in this Office action:
(a) A patent may not be obtained though the invention is not identically disclosed or described as set forth in section 102 of this title, if the differences between the subject matter sought to be patented and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art to which said subject matter pertains. Patentability shall not be negatived by the manner in which the invention was made.
The following are the references relied upon in the rejections below:
Zheng (Zheng, Mengyu, Chuan Zhou, Jia Wu, Shirui Pan, Jinqiao Shi, and Li Guo. "Fraudne: a joint embedding approach for fraud detection." In 2018 International Joint Conference on Neural Networks (IJCNN), pp. 1-8. IEEE, 2018.)
Chandola (Chandola, Varun, Arindam Banerjee, and Vipin Kumar. "Anomaly detection: A survey." ACM computing surveys (CSUR) 41, no. 3 (2009): 1-58.)
Epelman (US 2016/0210633 A1)
Pastore (US 2015/0371339 A1)
Claims 1, 3-6, 9, 11-14, 16 and 18-20 are rejected under 35 U.S.C. 103 as being unpatentable over Zheng and Pastore.
Regarding claims 1, 9 and 16, Zhang discloses a method, (and a related computer-readable storage medium and apparatus) comprising:
encoding, by an application executing on a processor based on an embedding function, transaction data for a first transaction…
P. 1, sec.(I), “Generally speaking, fraudsters always act in lockstep to increase total impact of target items. The fraud detection problem can therefore be viewed as finding suspicious dense blocks in the attributed bipartite graph, where the source nodes represent user accounts, the sink nodes are items, and the directed edges stand for feedback record. The attributes on each edge can be organized as a sensor composed of timestamp, rating score, review and so on.”
P. 3, first paragraph: “As shown in Fig. 2, the network embedding framework consists of two unsurprised components, i.e., source node representation part and sink node representation part.”
determining, by an embeddings layer of a neural network executing on the processor and based on the encoded transaction data for the first transaction, a vector for the first transaction, wherein the neural network is based on a plurality of positive entity pairs from a network graph of transaction data and a plurality of negative entity pairs not present in the network graph of transaction data, the negative entity pairs comprising artificially generated relationships between each entity in the negative entity pair;
P. 3, fig. 2 (reproduced below). The Examiner notes that the middle layer in each autoencoder neural network (flanking the words ‘local proximity’) is an embedding layer.
PNG
media_image1.png
412
640
media_image1.png
Greyscale
‘positive entity pairs’ :: P. 2 “Definition 5. (Local Proximity) The local proximity in a network is the local pairwise proximity between source node u and sink node v. For each pair of source-sink nodes linked by weighted edge, the number of reviews wu,v indicates the local proximity between user u and object v.”
‘negative entity pairs’ :: “If no edge is observed between source-sink nodes (u1 and v6 in Fig. 1), it means their local proximity is zero.”
processing, by the application, the first transaction based on the vector and an embedding space for the vector to determine one or more of:
(i) the first transaction is a fraudulent transaction, (ii) that the first transaction is associated with money laundering, or (iii) a recommendation based on the first transaction; and
P. 1, sec (I), “Online user feedback for the target items, such as reviews and ratings, usually incurs considerable influence for potential buyers. It is on this very basis that the fraudulent behavior has become more and more widespread. For example, one-third of consumer reviews and rates on the Amazon, and more than one-fifth of reviews on Yelp are estimated to be fake [1]. Users are easily cheated by these fake feedbacks and purchase items with poor quality. The user chum of ecommerce platform is then following. Hence, how to detect fraudsters and corresponding target items is a serious problem for both users and e-commerce platform.”
P. 2, “In order to detect fraudulent groups, we first propose a deep joint network embedding framework, named FraudNE, to embed source nodes u ∈ U and sink nodes v ∈ V jointly in one latent space. And then we make use of a clustering algorithm to find a high-density area in the latent space.”
transmitting, by the application, a result of the processing of the first transaction.
P. 5, algorithm 2, “Obtain fraudulent groups F”.
Pastore discloses the following further limitation which Zheng does not disclose:
the embedding function being a function that generates a vector representation of the first transaction based on an account ID, a merchant name, a location, and a purchase amount;
[0043] “Each payment card transaction record that is stored in the data warehouse 26 is associated with a consumer, and includes at least a date and time of the transaction, an account number and/or other identifying data of the cardholder making the purchase, a merchant ID and/or merchant name, and, generally, other merchant location and/or identification information of the merchant associated with the transaction, along with the financial details of the purchase.” (Emphasis added.)
At the time of filing, it would have been obvious to a person of ordinary skill to include the four specific features identified by Pastore in the fraud detection system of Zheng because each of these is potentially a useful feature for predicting fraud. In other words, each of these features may contribute to a more accurate predictive model. Both disclosure pertain to electronic payment processing.
Regarding claims 3, 11 and 18, Zheng discloses the following further limitation wherein the first transaction is associated with a first account, wherein processing the first transaction comprises:
clustering, by the application based on a model trained based on the neural network, the first transaction into a first cluster of transactions;
PP. 4-5, sec. (D) ‘Clustering’.
determining, by the application, that other transactions in the first cluster of transactions are associated with one or more accounts that have engaged in money laundering; and
P. 6, “Graph-based fraud detection methods often detect unexpectedly dense regions of the networks. Since spammers unavoidably generated edges in networks when they create fake reviews, these methods are hard to evade.”
determining, by the application based on the clustering of the first transaction into the first cluster of transactions and the determination that the other transactions in the first cluster of transactions are associated with one or more accounts that have engaged in money laundering, that the first transaction is associated with the money laundering.
P. 1, first col. “The fraud detection problem can therefore be viewed as finding suspicious dense blocks in the attributed bipartite graph, where the source nodes represent user accounts”.
P. 1, second col. “To address these issues, in this paper we propose an unsupervised method named FraudNE to detect abnormal users and items through deep joint network embedding.”
P. 2, “We then cluster groups of fraudsters and their corresponding abnormal items without the number of dense blocks as a priori. The main idea is shown in Fig. 1.”
Regarding claims 4, 12 and 19, Zheng discloses the following further limitation wherein the first transaction is associated with a first account, a first merchant, and a first location, wherein processing the first transaction comprises:
determining, by the embeddings layer of the neural network based on encoded transaction data for a second transaction, a second merchant associated with a second location; and
P. 1, first col. “Most attribute-based methods exploited the user accounts/items/feedbacks related features to address the suspicious dense blocks detection problem [2], [3]. However, these attributes-based methods for dense block detection are not adversarially robust. The fraudsters can fine-tune their text and behavior to make their features unsuspicious. For example, fraudsters can manipulate login times, their location, internet providers and IPs via large pools.”
generating, by the application for the first account, the recommendation specifying the second merchant associated with the second location.
P. 2, first col. “We then cluster groups of fraudsters and their corresponding abnormal items without the number of dense blocks as a priori.”
See also p. 5, algorithm 2, discussing cluster assignment.
Regarding claims 5, 13 and 20, Zheng discloses the following further limitation wherein the first transaction is associated with a first account, wherein processing the first transaction comprises:
determining, by the application, a similarity between the vector for the first transaction and a vector for a second transaction associated with a second account;
P. 5, algorithm 2, describing a clustering procedure. The Examiner notes that calculating similarity (distance vales between points / vectors) is inherent in every clustering procedure.
determining, by the application based on the similarity between the vector for the first transaction and the vector for the second transaction, that the first account is similar to the second account; and
Id. See description of cluster assignment (step 3).
generating, by the application based on the determination that the first account is similar to the second account, the recommendation comprising the second account.
Id.
Regarding claims 6 and 14, Zheng discloses the following further limitation wherein the application processes the first transaction based at least in part on a model, wherein the model is trained based on the embeddings layer of the neural network.
P. 1, fig. 1 (reproduced below), illustrating a node embedding layer.
PNG
media_image2.png
348
618
media_image2.png
Greyscale
Claims 2, 10 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Zheng, Pastore and Chandola.
Regarding claims 2, 10 and 17, Zheng discloses the following further limitation wherein the first transaction is associated with a first account, wherein processing the first transaction comprises:
determining, by the application, a distance in the embedding space between the vector for the first transaction and a vector for a second transaction, the second transaction associated with the first account;
P. 5, algorithm 2, step 1: “Find e neighbors of every point, and identify n core points with more than minPts neighbors.”
The Examiner notes that to identify neighboring points, an inter-point distance must be calculated. Thus, this feature is inherent in Zheng.
Chandola discloses the following limitations which Zheng/Pastore do not disclose:
determining, by the application, that the distance exceeds a threshold distance;
P. 22, “The concept of nearest neighbor analysis has been used in several anomaly detection techniques. Such techniques are based on the following key assumption: Assumption. Normal data instances occur in dense neighborhoods, while anomalies occur far from their closest neighbors.”P. 23, “The anomaly score of a data instance is defined as its distance to its kth nearest neighbor in a given data set. …. Usually, a threshold is then be applied on the anomaly score to determine if a test instance is anomalous or not.”
determining, by the application based on the distance exceeding the threshold distance, that the first transaction is the fraudulent transaction; and
Id.
See also p. 2, application to credit card fraud detection.
applying, by the application, an indication of the fraudulent transaction to the first account.
Id.
See also p. 2, “Anomalies in credit card transaction data could indicate credit card or identity theft [Aleskerov et al. 1997]”.
At the time of filing, it would have been obvious to a person of ordinary skill to combine the techniques disclosed by Chandola for anomaly detection (including fraud detection specifically) to the Zheng/Pastore system because—as noted by Chandola—data points which are far from its nearest clusters are anomalous, and may indicate fraud. Both Zheng and Chandola pertain to fraud detection.
Claims 7-8 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Zheng, Pastore and Epelman.
Regarding claim 7, Epelman discloses the following further limitation which Zheng/Pastore do not disclose wherein neural network comprising an embeddings layer, wherein the plurality of positive entity pairs are based on a respective time between a respective timestamp of the transactions of each positive entity pair being less than a time threshold.
[0083] “In an embodiment, the determination of an RTT value for the ZIP code pair (90210, 10111) may utilize these two records, assuming that they satisfy the “transaction pair” criteria specified by the particular configuration (e.g., travel time is less than a threshold, both transactions are of a particular type, etc.).”
At the time of filing, it would have been obvious to a person of ordinary skill to combine the technique disclosed by Epelman with the Zhang/Pastore system because it can provide additional information about fraudulent transactions that may be related, thus helping identify all related fraudulent transactions. Both Zhang and Epelman pertain to fraud detection.
Regarding claim 8, Zheng discloses the following further limitation wherein the network graph of transaction data is based on a plurality of transactions, the network graph of transaction data defining relationships between the plurality of transactions, each transaction associated with at least a merchant and one account of a plurality of accounts, the plurality of transactions excluding the first transaction.
P. 1, first col. “The fraud detection problem can therefore be viewed as finding suspicious dense blocks in the attributed bipartite graph, where the source nodes represent user accounts, the sink nodes are items, and the directed edges stand for feedback record. The attributes on each edge can be organized as a sensor composed of timestamp, rating score, review and so on.”
Regarding claim 15, Epelman discloses the following further limitation which Zheng/Pastore do not disclose the negative entity pairs comprising artificially generated relationships between each entity in the negative entity pair, the neural network comprising an embeddings layer, wherein the plurality of positive entity pairs are based on a respective time between a respective timestamp of the transactions of each positive entity pair being less than a time threshold.
[0083] “In an embodiment, the determination of an RTT value for the ZIP code pair (90210, 10111) may utilize these two records, assuming that they satisfy the “transaction pair” criteria specified by the particular configuration (e.g., travel time is less than a threshold, both transactions are of a particular type, etc.).”
The obviousness analysis of claim 7 applies equally here.
Additional Relevant Prior Art
The following references were identified by the Examiner as being relevant to the disclosed invention, but are not relied upon in any particular prior art rejection:
Ziegler discloses the application of neural network graph embeddings to financial transaction data. (Ziegler K, Caelen O, Garchery M, Granitzer M, He-Guelton L, Jurgovsky J, Portier PE, Zwicklbauer S. Injecting semantic background knowledge into neural networks using graph embeddings. In 2017 IEEE 26th International Conference on Enabling Technologies: Infrastructure for Collaborative Enterprises (WETICE) 2017 Jun 21 (pp. 200-205). IEEE.)
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Vincent Gonzales whose telephone number is (571) 270-3837. The examiner can normally be reached on Monday-Friday 7 a.m. to 4 p.m. MT. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Miranda Huang, can be reached at (571) 270-7092.
Information regarding the status of an application may be obtained from the USPTO Patent Center.
/Vincent Gonzales/Primary Examiner, Art Unit 2124